Texas Football Analytics (2015-2025)

A 6-notebook data science project analyzing Texas Longhorns football across betting markets, coaching eras, opponent-adjusted performance, recruiting pipelines, and in-game win probability - built on 10+ seasons of data from the College Football Data API.

No ESPN+ subscription. No proprietary scouting data. Just a free API and scikit-learn.

---

Why This Exists

College football analysis is dominated by hot takes and narrative. This project replaces opinion with data. It asks five concrete questions about Texas football and answers each one with a dedicated notebook, building from simple EDA up to a gradient boosting win probability model.

The through-line across all five analyses is coaching alpha - the gap between what a team's talent and schedule predict and what actually happens on the field. Texas has been one of the most talent-rich programs in the country for a decade, but talent alone does not explain outcomes. This project measures what does.

---

Key Findings

Analysis	Core Question
Betting (Project 1)	Is Texas consistently over- or undervalued by betting markets?
Coaching Eras (Project 2)	How do Strong, Herman, and Sarkisian compare across SP+, PPA, and recruiting?
Opponent-Adjusted (Project 3)	What happens when you control for schedule strength? Who generates real coaching alpha?
Recruiting (Project 4)	Does recruiting talent predict on-field success? How much variance is unexplained?
Win Probability (Project 5)	Can we model real-time win probability and measure clutch performance by era?

---

Notebooks

#	Notebook	Question	Method
0	project_0_texas_football_eda.ipynb	What does 10 seasons of Texas football look like?	EDA, scoring distributions, baseline ML
1	project_1_texas_betting_analysis.ipynb	Is Texas consistently over- or undervalued by betting markets?	ATS records, cumulative P&L, market inefficiencies
2	project_2_texas_coaching_eras.ipynb	How do Strong, Herman, and Sarkisian compare?	SP+ ratings, PPA, recruiting composites, Kruskal-Wallis
3	project_3_texas_opponent_adjusted.ipynb	What happens when you control for schedule strength?	Margin Over Expected, opponent-tier analysis
4	project_4_texas_recruiting_performance.ipynb	Does recruiting talent predict on-field success?	Lagged regression, transfer portal, NFL draft output
5	project_5_texas_win_probability.ipynb	Can we model real-time win probability and measure clutch?	Gradient boosting, calibration analysis, clutch delta

---

Coaching Eras

Era	Coach	Seasons	Conference	Context
Strong	Charlie Strong	2015-2016	Big 12	Inherited Mack Brown's program, struggled to sustain recruiting and results
Herman	Tom Herman	2017-2020	Big 12	Strong recruiting classes, 2018 Sugar Bowl season, inconsistent trajectory
Sarkisian	Steve Sarkisian	2021-2025	Big 12 / SEC	Rough 2021 debut, built to 2023 CFP contention, SEC transition in 2024

---

Quick Start

1. Clone and install

git clone https://github.com/lenamonj/texas-football.git
cd texas-football
pip install -r requirements.txt

2. Set your API key

Register for a free key at collegefootballdata.com.

export CFBD_API_KEY="your-api-key-here"

Or on Windows PowerShell:

$env:CFBD_API_KEY = "your-api-key-here"

3. Run

jupyter notebook

Notebooks are numbered 0-5 and designed to run in order, though each is self-contained with its own API calls.

---

Data Source

All data comes from the College Football Data API (CFBD). The free tier is sufficient.

Endpoint	Data
/games	Game results and box scores
/betting/lines	Historical betting lines across sportsbooks
/ratings/sp	SP+ advanced team ratings
/ratings/srs	Simple Rating System
/recruiting/teams	Team-level recruiting rankings
/talent	247Sports talent composite
/stats/season/advanced	PPA, success rates, advanced team stats
/player/returning	Returning production metrics
/draft/picks	NFL draft selections
/play/stats	Play-by-play data for win probability

---

Design Decisions

• Temporal integrity - All models respect time. No future data leaks into historical analysis. Recruiting is lagged 1-3 years because that is when those players actually contribute.
• Margin Over Expected (MOE) - Raw win-loss records are misleading without schedule context. MOE = actual margin - SP+ predicted differential isolates coaching effect from talent and schedule noise.
• Gradient boosting over logistic regression - The win probability model uses GBM because score-time interactions are inherently nonlinear. Logistic regression baseline included for comparison.
• Kruskal-Wallis over ANOVA - Game margins are not normally distributed. Nonparametric tests are the correct choice for comparing coaching eras.
• Conference transition handled - Texas moved from Big 12 to SEC in 2024. Conference-level splits account for this transition rather than treating all seasons as one conference.
• Clutch delta as the key metric - Win rate in close Q3-Q4 situations minus model-predicted win rate. This connects the betting analysis to the coaching analysis and explains where alpha actually comes from.

---

Project Structure

texas-football/
|-- README.md
|-- LICENSE
|-- requirements.txt
|-- .gitignore
|
|-- project_0_texas_football_eda.ipynb           # Exploratory analysis, baseline ML
|-- project_1_texas_betting_analysis.ipynb       # ATS performance, over/under, P&L
|-- project_2_texas_coaching_eras.ipynb          # Strong vs Herman vs Sarkisian
|-- project_3_texas_opponent_adjusted.ipynb      # Schedule-controlled MOE
|-- project_4_texas_recruiting_performance.ipynb # Talent pipeline, portal, draft
+-- project_5_texas_win_probability.ipynb        # In-game GBM model, clutch analysis

---

Dependencies

Package	Purpose
pandas	Data manipulation and time series
numpy	Numerical computation
matplotlib	Visualization
seaborn	Statistical plots
scikit-learn	Gradient boosting, logistic regression, evaluation metrics
scipy	Kruskal-Wallis, Mann-Whitney U statistical tests
cfbd	College Football Data API Python client
requests	HTTP requests to CFBD API

---

License

This project is licensed under the MIT License.

---

_{Built with Python and the College Football Data API. No proprietary scouting tools required.}